Food manufacturing foreign divestment and domestic investment in developed countries

Whilst there is some literature on the effect of inward foreign direct investment on domestic investment for the whole economy and the agricultural sector, that of foreign divestment on domestic investment for food manufacturing is rare. This paper contributes to the literature by estimating the crowding effect of foreign divestment on domestic investment in the food manufacturing sector using an unbalanced panel of 29 countries from 1991 to 2019. Foreign divestment crowded out domestic investment for developed countries in the short and long runs. In terms of the absolute reduction in domestic investment, the short-run effect is higher than the long-run effect. Policies to attract inward foreign direct investment and retain it should be pursued.

the relationship between costs of capital (interest rate) and investment is negative. This is necessary for establishing a stable full-employment general equilibrium [73]. Variants of the neoclassical theory emphasise adjustment between current and optimal capital stock. The adjustment has associated costs. These could be convex, increasing at an increasing rate in DI or non-convex [74][75][76].
Barrios et al. [77] and Wang [72] theorised that the entry of foreign firms into a host country could increase the population of domestic firms initially. The competition in the final product market would lead to price competition. This could cause domestic firms to exit the market. Where some firms produce inputs for others, the population of domestic firms would rise. In line with this theory, FD could either decrease the population of domestic firms or increase it. The resultant effect is an increase or decrease in DI. This outcome is subject to empirical examination.

Empirical review
The empirical review focuses on the relationship between total manufacturing private investment and total DI on one hand, and inward foreign direct investment, inflation, and growth on the other hand. Afful and Kamasa [32], Baltar et al. [32], Ebghaei [78] and Panda and Nanda [33] studied manufacturing private investments in Ghana, Brazil, Iran, and India. All except Panda and Nanda [33] applied Autoregressive distributed lag (ARDL) models whilst Panda and Nanda [33] used the general method of moments (GMM) estimator.
Afful and Kamasa [31] reported a significant negative relationship between inflation and investment in Ghana. However, Nguyen [39] found a significant positive relationship in the case of developed countries and no significant effect of inflation on investment in developing countries. The former explained that increasing prices reflect a fall in the value of the currency, hence declined desire to save in the banks resulting in a fall in investment. Exchange rate was found to be negatively related to DI in the US but neutral for France, Germany, Japan, and UK (Bahmani-Oskooee & Baek [35]).
Regarding economic growth, Afful and Kamasa [31] and Panda and Nanda [33] concluded that growth positively influenced investment in the manufacturing sector in Ghana and India, respectively. The latter noted that investment increases during periods of boom and are likely to fall during periods of recession. Economic growth positively influences investment in -France, Germany, the UK, and the US (Bahmani-Oskooee and Baek, [35]). Boateng et al., [38]; Djokoto, [27]; Farla et al. [7], and Nguyen [39], found positive effect of economic growth on DI in developing countries. Bakari [26] and Bakari et al. [37] however, found no significant relationship between economic growth and DI in France and Germany based on an ARDL estimation. For Japan, Bahmani-Oskooee and Baek [35] also reported no significant effect. There were no attributions for these findings.
Three findings of Djokoto [40], the only study on the effect of foreign direct investment on domestic investment in the manufacturing sector, are relevant here: First, developed economies experienced a crowd-out effect of FDI on DI in the short run, whilst the others experienced no significant effect. Second, food manufacturing sectors of developed, developing and transition economies exhibited a substitution effect in the long run. Third, the growth of food manufacturing value-added positively influenced domestic investment in food manufacturing.
It would be observed from the empirical review that only one study addressed the determinants of DI for food manufacturing. However, this did not address foreign divestment. This paper fills this void.

Data
Foreign divestment is less common and somewhat ignored compared to FDI inflow or outflow 22,78. Thus, there were only 149 observations on FD records on food manufacturing in 29 developed countries (Appendix) from 1991 to 2019 in FAOSTAT [13]. Proceeding with the 149 could reduce the efficiency of the estimates. Thus, data on food manufacturing inward FDI (IFDI) for developed countries was used. The 149 observations of the FD were isolated using interaction with a dummy variable. Consequently, the number of observations increased from 149 to 518 for the 29 countries from 1991 to 2019. The span of the data as well as the developed countries in the data depended solely on the availability of data. The resulting observations are large enough to produce efficient estimates. The panel was unbalanced and there was no need to supply missing values as the key variable of FD must occur and must not be computed.

Model and modelling
The Agosin and Machado [29] model has been popular in investigating the effect of FDI on DI [40,[79][80][81][82]. However, the model has some limitations as it was developed based on assumptions for developing countries. Further, the only control variable is the economic growth rate. An alternative approach, which accounts for both outward and inward FDI simultaneously and additional control variables that are known to explain DI, is used in this study [83][84][85][86][87][88]. The alternative approach accounts for other covariates of the foreign direct investment. Therefore, equation (1) is specified.
Where i is the number of cross-sections (countries) and t is the period. α k , where k = 1, 2 … 9, are parameters to be estimated. DI is the domestic investment, L1.DI is one year (period) lag of DI. DI_FD is the interaction between DI and FD. FD is foreign divestment = 1 and zero otherwise. At the aggregate level, the FD of IFDI is negative of IFDI [49,51]. It is worthy of note that, UNCTAD [89], the source of the FAOSTAT data, loans to foreign affiliates are subtracted before arriving at the reported values on a directional basis captured as FDI (outward or inward). Indeed, the definition of FD as the negative of IFDI or OFDI can be found in the United Nations literature [90][91][92]. IFDI is inward foreign direct investment whilst IFDI_FD is the interaction of FD and IFDI. OFDI is an outward foreign direct investment and FBTTO is trade openness (imports plus exports divided by food manufacturing GDP). The former is necessary as it occurs together with IFDI especially, in developed countries. Food manufacturing firms import raw materials and export finished and semi-finished goods. Trade provides resources for production and market access for food manufacturing output. These must have some influence on investment. INFLA is the annual growth rate of the consumer price index. Increased inflation decreases the purchasing power of consumers. This would reduce consumption and increase the stock of food manufacturers. This will in turn slow down investment in food manufacturing. GR is the annual growth rate of food manufacturing GDP. Increased economic output would increase income for both consumption and savings. These would increase consumption and reduce the stock of manufactured goods thus, triggering additional investment in food manufacturing. SR is the savings rate (savings to national GDP ratio). Theoretically, the interest rate explains investment. That is, a decrease in lending interest rates will encourage investment. Simultaneously, a decrease in interest rate on savings will decrease savings and consequently, loanable funds. Thus, interest rate mediates savings and investment. Apart from INFLA, GR and SR, all other variables are expressed in terms of food manufacturing GDP. INFLA and SR, obtained from the World Development Indicators of the World Bank, are for the total economy as there were no corresponding series for food manufacturing. This is not out of place as the DI of food manufacturing does exist in the economy and is affected by these indicators of the total economy. The savings rate was substituted for the interest rate as there were inconsistencies in the interest rates over the period. DI, IFDI, OFDI, and FD were extracted from FAOSTAT [13]. FBTTO and GR were calculated using data obtained from FAOSTAT [13] and UN Commodity and Trade Statistics (https://comtrade.un.org/).
From equation (1), the effects of FD on DI are computed based on the combination of coefficients in Table 1. These are computed as the Wald and tested by the chi-square test statistic. The application of the chi-square test of the Wald statistic for the long run crowding effect is as follows: a. Failure to reject the null hypothesis that η LR =1 implies there is no crowding effect. b. Rejection of the null hypothesis, η LR = 1 and η LR >1, is evidence of a crowding-in (complementary) effect. c. Finally, rejection of the null hypothesis, η LR = 1 and η LR < 1, is evidence of a crowding-out (substitution) effect.In the short run d. Failure to reject the null hypothesis that η SR = 1 means there is no evidence of a crowding effect. e. If the null hypothesis, η SR = 1 is rejected and η SR >1 then, this is a crowding-in (complementary) effect. f. In case the hypothesis, η SR = 1 is rejected and η SR <1 then, it is a situation of crowding-out (substitution) effect.

Estimation procedure
The 29 cross-sections (countries) exceeds the mean period of 16 years. Thus, the cross-sectional characteristics outweigh the time series characteristics in the model [93][94][95][96]. Following the data structure, panel fixed, and random effects estimations were obtained. The Hausman test [97] was used to choose between fixed effects (FE) and the random effects (RE) estimations. Endogeneity could not be ruled out from the data, therefore, endogeneity was accounted for by estimating the general method of moments (GMM). The collapse option was used in the estimation of the GMM to control for instrument proliferation.

Background of data
The mean DI is 0.2213 of food manufacturing GDP coinciding with Austrian data for 2010, with a minimum of 0.1061 (Greece, 2019) and a maximum of 0.5360 (Ireland, 2019) ( Table 2). The mean IFDI is 0.0325. The minimum is − 0.4209, the same as IFDI_FD. OFDI also experienced divestment to the tune of 3.0605 of food manufacturing GDP. The mean savings rate is 24% of the national GDP with a minimum of 7.5% (Greece, 2012) and a maximum of 58.8% (Ireland, 2019). On average, food manufacturing value added grew by 3.19% against a minimum of − 67% reported by Romania in 2003 and a maximum of 47.3% reported by Romania in 2008. Table 3 reports the estimations of fixed and random effects. Both models 1 and 2 show heteroskedasticity and the existence of serial correlation. The null hypothesis of the Hausman test is rejected. Thus, the FE model is preferred to the random-effects model. Model 1 Table 1 Computation of effects.

Effect
Computation Null hypotheses Short run (η SR ) Long run (η LR ) Note. The -chi-square test is applicable.  Table 4. Not only are the estimates of the key variables similar across models 3 to 9, but the estimates of the control variables in models 4-9 are also like those of model 3. This is evidence of the robustness of the estimates of the key variables to the control variables. Indeed, the presence of the control variables did not significantly influence the estimates of the key variables.
Whilst it was appropriate to employ the linear panel estimators, FE and RE, endogeneity between the explained and the explanatory variables could be present. To account for this, the GMM estimator was applied to the data (Table 5). In model 10, the number of instruments was 63, far above the number of cross-sections, 29. Thus, the collapse option was used to prevent the proliferation of the instruments in the GMM estimation. The result is model 11. The probability of the second-order serial correlation is statistically insignificant. Whilst model 10 has more statistically significant estimates than model 11, the magnitude of the estimates of models 10 and 11 are generally alike. It is worth noting that the probability of the Sargan statistics in model 10 is 0.9995 whilst that of model 11 is 1.0000. Thus, controlling for instrument proliferation did not reduce the level of probability of the Sargan statistics. As both levels of probability are greater than 0.10, this suggests that the overidentifying restrictions assumed in the estimation procedure are valid. The standard errors of model 11 are finite-sample corrections as the default two-step standard errors are biased in finite samples due to the neglected sampling error in the weighting matrix based on Windmeijer's [97] approach.
Another robustness to control variables was carried out, this time, with the GMM for model 11. The results in Table 6 show that the estimates of the key variables are similar across models 11 to 17. Also, the estimates of the control variables in models 12-17 are like  Table 3 Panel linear estimations.  those of model 11. Therefore, the estimates of model 11 are robust to control variables. The probability of the second-order serial correlation test and the Sargan test is also appropriate.
To aid the discussion and compare the panel linear estimations to the GMM estimates, models 3 and 11 are re-presented in Table 7. Although model 3 possesses more statistically significant variables than model 11, the estimates are generally alike in sign and magnitude. This also suggests robustness between the POLS estimates and the GMM estimates. Therefore, the results obtained from fitting equation (1) to the data are robust to estimator, control variables and violations of the assumptions of linear regression.

Discussion of other variables
The coefficients of IFDI are positive and statistically significant. This suggests IFDI has no discernible effect on food manufacturing DI in the short run for developed countries. This is a departure from the findings of Djokoto [40] for developed countries. The  Notes: 1. Values in parenthesis are Driscoll-Kraay standard errors. 2. ***,**,* are 1%, 5% and 10% levels of statistical significance respectively.

Table 5
General method of moments (GMM) estimations. departure may be due to the reason that the model of Djokoto [40] contained fewer control variables than this study. Indeed, the inclusion of statistically significant control variables not only draws from the variability in the dependent variable but may also reduce the statistical significance of the other covariates.
The coefficients for OFDI are positive and statistically significant for model 11. One United States dollar increase in OFDI will induce a US$0.02 increase in food manufacturing DI in developed countries. This is a small effect, though. The positive sign can be attributable to joint investments of MNEs, both at home and abroad. Some goods from foreign subsidiaries could be inputs for the home finite-sample correction as the default two-step standard errors is biased in finite samples due to the neglected sampling error in the weighting matrix.

Table 7
Robustness to endogeneity (fixed effects and GMM) estimations. factory, hence, an increase in investment abroad would lead to an increase in home investments as well.
The coefficients of food manufacturing trade openness (FBTTO) and inflation (INFLA) are statistically insignificant. Thus, these do not have a discernible effect on DI. Whilst the statistical insignificance of the coefficient departs from Afful and Kamasa [31], the sign concurs with their finding.
The coefficients of GR, although negative for both models, it is statistically significant for model 3 and statistically insignificant for model 11. It was expected that an increase in GR should spur DI, however, this did not turn out to be the case. Whilst this result agrees with those of Bahmani-Oskooee and Baek [35], Bakari [36] and Bakari et al. [37] respectively for Japan, France and Germany, other studies have reported positive and significant coefficients for the growth rate on private investment [7,31,33,38,39]. It must be noted that the value added is for food manufacturing. Resources from food manufacturing are inadequate for investment in food manufacturing. Additional and perhaps more substantial savings would arise from other sectors to fund investment in food manufacturing hence, the negative and statistically insignificant sign.
The coefficients of SR are positive and statistically significant for both models. A one US dollar increase in savings would induce 21 cents increase in domestic investment in food manufacturing in developed countries. The positive sign is in line with the theory that resources from households or surplus units become investable funds for industry or deficit units. Recalling that the savings rate is for the whole economy whilst the DI is for only food manufacturing, the 21 cents suggest that about 80 cents of the US$1.00 saved are channelled into sectors other than food manufacturing. The 21 cents is about 10 times the share of food manufacturing in GDP in developed in 2017 [5].

Crowding effect of foreign divestment on domestic investment
Except for L1.DI, the coefficients of all key variables are not statistically significant. The statistical significance of L1.DI implies previous year's DI would influence the current year's DI. Aside from conforming to theoretical expectations, the positive and statistical coefficient is particularly useful for GMM estimation.
The application of the Wald in Table 1 is based on estimates in models 3 and 11 and is presented in Table 8. The Wald from both the fixed effects and GMM estimations is less than 1.00. The test of difference in the Wald suggests that these are statistically not different. As the GMM accounts for endogeneity, as is common with macroeconomic variables, we focus on the GMM effects. In the short run, the null hypothesis that the Wald is equal to 1.00 is rejected. Specifically, the Wald is less than 1.00. This implies that a US$1.00 increase in FD would increase DI by US$0.1631. This shows that as much as US$0.84 (1-0.1631) of domestic investment is lost due to an increase in FD of US$1.00. This is certainly a substitution or crowding-out effect of FD on DI, in the short run.
The null hypothesis that the Wald for the long run is equal to 1.00 is also rejected. Specifically, in the long run, a US$1.00 increase in FD would induce a US$0.66 increase in DI. This implies that US$0.34 is lost in DI owing to an increase in FD. This is also a crowdingout effect or substitution effect. Thus, in the long run, FD crowds out (substitutes) DI in food manufacturing in developed countries. The crowding-out effect is unsurprising because domestic investment includes all investments in the host economy. Thus, a reduction in investments via FD would reduce DI. As all three forms of FD, downsizing, relocation, and termination [17][18][19][20] result in the loss of IFDI from the host economy, the substitution (crowding-out) effects can be explained.
It would be observed that the magnitude of the increase in DI due to FD is higher in the long run than in the short run. That is, the substitution effect declines as one moves from the short run to the long run. In the short run, FD comes as a shock. However, in the long run, firms that have downsized could increase investments. For firms that have relocated and/or terminated their operations, firms in the host economy could acquire some resources including space and operate similar or other businesses in food manufacturing. As in the case of Wald, the monetary values of the substitution effects have a like pattern.
Some other reasons can be adduced for the reduction in DI. First, FD creates a loss of business and investment for suppliers and offtakers of foreign-divested firms. Second, the loss of jobs and income to households would lead to low savings hence reduced investment. Third, FD could lead to reverse technology transfer. Fourth, FD could lead to a reduction in exports of food manufacturers in the case of export-oriented foreign food manufacturing firms. Fifth, the decline in exports not compensated for from other sectors could lead to increased imports with associated foreign exchange pressures and imported inflation. This can dampen wealth levels and reduce savings and hence investments. The sixth reason is that the reasons for divestment may themselves be a disincentive to firms intending to invest in the economy, both local and foreign hence, would induce a decline in DI. Thus, their accentuation would lead to the loss of DI. Finally, the seventh is that these causes could have a multiplier effect on the economy which would further drive down DI.

Conclusions and recommendations
Whilst there is some literature on the effect of inward FDI on DI for the whole economy and the agricultural sector, FD on DI for the whole economy, agriculture, and food manufacturing, is minimal. This paper contributes to the literature by estimating the crowding effect of food manufacturing FD on food manufacturing DI. Further, this paper computed the changes (reduction) in DI occasioned by the FD.
FD crowded out DI for developed countries in the short and long run. In terms of absolute reduction in DI, the short run experienced a greater reduction than the long run. The FD thus, appear to be a trigger for a reduction in DI.
These conclusions necessitate some recommended actions. First, FDI regulatory agencies should be responsive to the concerns of food manufacturing foreign investment firms. Second, where FD becomes inevitable, the agencies should require MNEs in the host country to provide advance notice to the agency on their divestment decisions. Aside from another opportunity to assist the MNEs, the notice could be circulated by the agency so that domestic or foreign firms could acquire or take over these firms. Third, the macroeconomy should be improved as the factors that promote FD could discourage IFDI. Policymakers and economic managers in developed countries must target to increase domestic investment in food manufacturing by at least US1.7b annually to make up for the loss due to FD.
Further research can consider developing countries as these tend to require more investment than developed countries and FD would be a greater disservice to them than developed countries.